Commenced in January 2007
Frequency: Monthly
Edition: International
Paper Count: 72667
Interpretation of Cone Penetration Test’s Tests to Characterize Tropical Soils With Recourse of Machine Learning

Authors: Jeniffer Viegas, António Gallardo, Lucas Bottaro, Rodrigo Marinaro


Cone penetration test (CPT) has been strongly applied to identify the soil profile and to provide some estimation of soil parameters. Several correlations exist, allowing the geo-characterization of the soil from CPT data, such correlations must be carefully applied, and whenever possible, corrected with laboratory tests. Tropical soil has an inherent variability capable of providing very distinct results from very similar samples. Project designers must deal with this variability, and correctly characterize these materials. The present work focuses on a case study where the goal was to distinguish and characterize two soft soils placed on the foundation of a tailings dam in Brazil. The dam is still in construction, and its foundation belongs to a complex geological environment with soft soils that can reach Nspt blow as low as its own weight. The geological survey identifies two horizons of soft soils: i) saprolite of dolomitic phyllite and ii) residual soil of saprolite of dolomitic phyllite. However, spatially distinguishing these two soils has shown to be a challenging task. They are quite similar in most parameters, and from lab tests, the parameter that helps separate these soils is the pore pressure Skempton parameter – A. The groundwater level in this place is not clear also, which difficult its estimate to interpret the vertical effective stress profile and further parameters from the CPT analysis. To overcome this issue a sensitive analysis of the influence of groundwater level on the parameters of interest in this work (Overconsolidation ratio and undrained strength ratio) was performed. To get as much information as possible from all datasets available, an Exploratory data analysis (EDA) followed by the application of an unsupervised learning algorithm was accomplished. Although an exactly spatial division from these soils were not possible, the EDA and unsupervised learning allow better visualization of the spatial distribution of these soils and grouping by desired characteristics, such as the pore pressure parameter.

Keywords: tropical soils, CPT analysis, machine learning, geotechnical characterization

Procedia PDF Downloads 17